anchor
Findings
- IC-1395ChatGPT's ethical violation rate decreases from 70.07 to 57.58 APV when given targeted in-context value instructions generated by VILMO, outperforming baseline alignment methods [supporting]
- IC-1518Domain finetuning of LLaMA 2 7B, LLaMA 2 13B, and GPT-2 XL on PubMed causes topic and style priors to shift dramatically, accounting for the majority of the probability change, while factual knowledge learning contributes only a small fraction [supporting]
- IC-1519Topic and style biases in LLaMA 2 7B are learned like simple features (rapidly, with minimal capacity, concentrated at the first few tokens, magnified by learning rate) while factual knowledge is learned like complex features (slowly, requiring significant capacity, uniformly across positions, unaffected by learning rate) [primary]
- IC-1580The official Llama 2-7B checkpoint fails to generate valid numerical responses for 3D-dependent molecular properties, with a valid answer rate of only 23% for SCF energy [supporting]
- IC-248Instruction fine-tuning causes context reliance under knowledge conflicts to initially increase then decrease (context-parametric inversion) in Llama2-7B, Pythia-6.9B, and Mistral-7B [supporting]
- IC-282GPT-2 XL (1.5B) exhibits lower accuracy but reduced overconfidence (smaller ECE and Brier scores) compared to larger models on the CAT benchmark [supporting]
- IC-394Text generation in SDXL, DeepFloyd IF, and SD3 is controlled by less than 1% of parameters concentrated in specific cross- or joint-attention layers, and these layers are specialised for text content rather than visual template [supporting]
- IC-409Knowledge editing methods correct verified hallucinations in Llama2-7B, Llama3-8B, and Mistral-v0.3-7B far less effectively than their scores on existing benchmarks suggest [primary]
- IC-410Knowledge editing can degrade generalization performance below pre-edit levels in Llama2-7B, Llama3-8B, and Mistral-v0.3-7B [primary]
- IC-411Llama2-7B, Llama3-8B, and Mistral-v0.3-7B do not reason with edited knowledge in multi-hop questions, as editing methods mostly underperform pre-edit portability scores [primary]
- IC-412Edited knowledge in Llama2-7B is significantly less robust to adversarial prompts than in Llama3-8B and Mistral-v0.3-7B [primary]
- IC-569Mistral-7B-instruct-v0.1 achieves only F1 of 0.419 on zero-shot stance detection for the X-Stance German dataset, substantially below the fine-tuned BERT baseline (F1 0.693) [supporting]
- IC-645CLIP, PickScore, and HPSv2 text embeddings share a common direction (cone effect) that captures text-irrelevant preferences, and the orthogonal component c⊥p better measures T2I alignment; CLIP's untrained common direction makes it ineffective for reward fine-tuning [supporting]